Real-world multivariate time series often exhibit multi-scale temporal dynamics and intricate inter-variable dependencies, making long-term forecasting particularly challenging. In this work, we propose a global token-driven multi-scale forecasting framework with dual-attention fusion. To capture multi-scale periodic patterns, the input sequence is segmented into multi-scale patches based on candidate periods, with the dominant ones derived via fast Fourier transform. Global tokens are then introduced as shared representations to integrate information from the temporal and variable dimensions, and a multi-scale patch-token interaction module is designed to establish interactions between the patches and global tokens, enabling the capture and aggregation of temporal dependencies across different scales. A dual-attention fusion module, employing both self-attention and cross-attention mechanisms, is then proposed to capture intrinsic and context-aware variable correlation among variables. To integrate cross-variable and cross-scale information into patch representations, a global information fusion module is designed. Finally, a period-aware weighting approach is devised to adaptively fuse the multi-scale predictions. Comprehensive experiments demonstrate that the proposed framework achieves state-of-the-art performance across various real-world datasets.

Effective task-oriented semantic communications relies on perfect knowledge alignment between transmitters and receivers for accurate recovery of task-related semantic information, which can be susceptible to knowledge misalignment and performance degradation in practice. To tackle this issue, continual knowledge updating and sharing are crucial to adapt to evolving task and user related demands, despite the incurred resource overhead and increased latency. In this paper, we propose a novel collaborative knowledge sharing-empowered semantic transmission mechanism in a two-tier edge network, exploiting edge cooperations and bit communications to address KB mismatch. By deriving a generalized effective semantic transmission rate (GESTR) that considers both semantic accuracy and overhead, we formulate a mixed integer nonlinear programming problem to maximize GESTR of all mobile devices by optimizing knowledge sharing decisions, extraction ratios, and BS/subchannel allocations, subject to task accuracy and delay requirements. The joint optimum solution can be obtained by proposed fractional programming based branch and bound algorithm and modified Kuhn-Munkres algorithm efficiently. Simulation results demonstrate the superior performance of proposed solution, especially in low signal-to-noise conditions.

Despite its popularity, federated learning (FL) in heterogeneous networks faces two critical challenges, i.e., the straggler problem due to devices with limited capabilities and low resource utilization rate of the FL server. The straggler problem arises when devices with limited computational capabilities delay the convergence of the global model. On the other hand, the computational resources of the FL server are often underutilized, mainly due to its relatively simple involvement for model aggregation. To tackle the issues in diverse scenarios, we propose a new joint computational resource allocation and layer partitioning (JCRALP) scheme to improve the overall FL performance by leveraging the capabilities and resources of both FL server and all clients. In the scenario where system parameters regarding the computational capabilities of the clients and the task burden can be accurately measured, we propose an optimization-based approach that leverages our proposed multi-step water-level equalization algorithm and the incremental ceiling adjustment algorithm. In the scenario where parameters cannot be measured accurately, we propose a reinforcement learning-based method using a modified twin delayed deep deterministic policy gradient algorithm. Extensive simulation results demonstrate that JCRALP efficiently and effectively mitigates the straggler problem and inclusively enables more client participation in FL. By including more datasets, the global model becomes more representative, while server computational resources are utilized more efficiently, significantly reducing convergence latency.

In task-oriented semantic communications, the transmitters are designed to deliver task-related semantic information rather than every signal bit to receivers, which alleviates the spectrum pressure by reducing network traffic loads. Effective semantic communications depend on the perfect alignment of shared knowledge between transmitters and receivers, however, the knowledge alignment cannot always be guaranteed in practice. In multi-cell networks, due to heterogeneous transceivers with distinct knowledge bases and limited computation capabilities, and random channel conditions in between, it is challenging for mobile devices (MDs) to access the best small base station (SBS) to perform effective semantic communications and complete requested tasks. To address the knowledge mismatch issue, we propose a novel task-oriented semantic transmission mechanism, leveraging knowledge sharing and bit communications to guarantee the effective target task execution. To maximize the derived semantic-based performance metric, i.e., generalized effective semantic transmission rate of all MDs under the designed mechanism, a mixed integer nonlinear programming problem is formulated to jointly optimize knowledge sharing decisions, semantic extraction ratios, and SBS associations while satisfying the semantic accuracy and delay requirements of target tasks. By decomposing the formulated problem into multiple subproblems equivalently, an optimum algorithm is proposed and another efficient algorithm is further developed using hierarchical class partitioning and monotonic optimization. A variety of simulation results demonstrate the validity and excellent performance of proposed solutions over a wide range of system parameters.

This paper introduces algorithms for static task class partitioning in mobile computation offloading (MCO). The objective is to partition a given set of task classes into two sets that are either executed locally by the mobile device (MD) or those classes that are permitted to contend for remote edge server (ES) execution. The goal is to find the task class partition that gives the minimum mean MD power consumption subject to task completion deadlines. The paper generates these partitions for both soft and hard task completion deadlines. Two variations of the problem are considered. The first assumes that the wireless and computational capacities are given and the second generates both capacity assignments subject to an additional resource cost budget constraint. The proposed partitioning algorithms are based on heuristic class ordering methods. The paper introduces two class ordering methods, a simpler one based on a task latency criterion, and an hierarchical version that first sorts and groups classes based on a mean power consumption criterion and then orders the task classes within each group based on a task completion time criterion. A variety of simulation results are presented that demonstrate the excellent performance of the proposed solutions for both given and optimized network resource assignments.

Digital twins (DTs) can be used to represent the behavior of real physical systems (PSs) in their interaction with other objects. Each DT periodically communicates with its PS and uses these updates to implement features that reflect the real behavior of the PS. A given feature can be implemented using different models that create the feature with differing levels of system accuracy. In this article, we study the DT model selection problem, where the DTs of multiple PSs are hosted at an execution server (ES). The objective is to maximize the minimum feature accuracy for the requested features by making appropriate model selections subject to the synchronization and ES execution constraints. The model selection problem is first formulated as an NP-complete integer program. It is then decomposed into multiple subproblems, each consisting of a modified Knapsack problem. A polynomial-time approximation algorithm is proposed using dynamic programming to solve it efficiently, by violating its constraints by at most a given factor. A generalization of the model selection problem is then given and an approximation algorithm using relaxation and dependent rounding is proposed to solve the problem efficiently with guaranteed constraint violations. A variety of simulation results are presented that demonstrate the excellent performance of the proposed solutions.

In mobile computation offloading (MCO), mobile devices (MDs) can choose to either execute tasks locally or have them executed on a remote edge server (ES). This paper addresses the problem of assigning the wireless communication bandwidth and the ES capacity used for the task execution, so that task completion time constraints are satisfied. The objective is to minimize the average power consumption of the mobile devices, subject to a cost budget constraint for obtaining the communication and computation resources. The paper includes contributions for both soft and hard task completion deadline constraints. The problems are first formulated as mixed integer nonlinear programs (MINLPs). Approximate solutions are then obtained by decomposing the problems into a collection of convex subproblems that can be efficiently solved. Results are presented that demonstrate the quality of the proposed solutions, which can achieve near optimum performance over a wide range of system parameters.

Digital twins (DTs) are virtual implementations of real physical systems (PSs) that interact with other objects on their behalf. Each PS periodically communicates with its digital twin so that the state of the DT is always sufficiently current. Using these updates, a DT can provide features that represent the real behavior of its PS using models that yield differing levels of system accuracy. In this paper, we study the DT model selection problem in wireless networks where the DTs of multiple PSs are hosted at an edge server (ES). The accuracy obtained from a given model is a function of its required amount of PS input data, the updating frequency, and the amount of computational capacity needed at the ES. The objective is to maximize the minimum achieved accuracy among the requested features by making appropriate model selections subject to wireless channel and ES resource availability. The problem is first formulated as an NP-complete integer program. The paper then uses relaxation and dependent rounding, and introduces a polynomial time approximation algorithm to obtain good solutions. A variety of simulation results are presented that demonstrate the excellent performance of the proposed solution.

This paper proposes a method of joint wireless network and job service allocation for use with mobile computation offloading where task completion times have deadline constraints. In this design, mobile devices (MDs) may execute a computational task locally or offload the task through a wireless network for execution on an edge server (ES). The network owner offers to lease wireless communication channels at a given set of base stations along with edge server capacity that is used for job execution. The objective is to obtain a wireless and service capacity allocation that minimizes the total energy consumption of the mobile devices, subject to a cost budget constraint and constraints on the delay incurred by offloaded task execution. The design is first formulated as a mixed integer nonlinear programming problem. An approximate solution is then obtained by decomposing it into a collection of convex subproblems that can be efficiently solved. Results are presented that demonstrate that the proposed solution achieves near optimum performance over a wide range of system parameters.